| With the rapid development of science and technology,driverless technology,has emerged from the beginning of the century,is becoming increasingly mature after more than ten years of evolution.In recent years,the researchers started from the theory,and put forward excellent and outstanding reasonable methods in the sensor selection,processing type,technical algorithm and other aspects.The first task of the research is to ensure the safety of the driver,so object detection technology is highly valued.Obstacle detection is a more difficult part of it,because the driving assistance system is difficult to estimate whether the target is obstacles.Although the current detection technology can detect obstacles around cars and pedestrians on the road,it cannot be used as obstacles in unknown and complex objects.In view of the above problem that the obstacle is complex and difficult to detect,this paper presents a small and unknown obstacles detection technology based on around view system in the parking.The essence of the detection method is based on the camera imaging principle and the nonlinear feature of the perspective transformation.That is,the motion of the obstacle in the image sequence in the panoramic view causes different displacement because of height,and the obstacle is detected by comparing the displacement with the car.Firstly,the fisheye lens were placed around the car,before calibrating and correcting them.Four images of the same moment were converted to the panoramic view and composed a panoramic image.Secondly,presenting a model detecting the intersection of the parking line between adjacent frames.Roughly extracted the right-angle point of the outline of the white area.According to the distance between the parking line intersection point,the connection angle and other standards for screening.Taking the width of the parking line as the distance to find the paired intersection point,matched based on the Euclidean distance.After obtaining the ground feature points in two adjacent frames,self-vehicle motion estimates was calculated by the twodimensional ICP algorithm for the simulated current frame of the previous frame,and eliminating the dynamic background.Finally,proposing a method of obstacle detection based on the HSV space.Subregions were obtained by dividing the hue channel into six intervals for image color segmentation.Comparing the corresponding subregions of the current frame with the simulated frame.If the subregion belonged to the ground,the subregions of both frames were the same.On the contrary,subregion belonged to obstacles,they had different subregions due to unequal displacement.By comparing the position and size of the external rectangle of the contours of subregions,the obstacle was detected finally.To verify the feasibility of the detection algorithm,putting various small obstacles in the parking lot.Six video sequences were shot according to the different types of obstacles.Video sequence 1 and 2 contained road cones,poles and other static obstacles,video sequence 3 and 4contained bicycles,pedestrians and other dynamic obstacles,and video sequence 5 and 6contained garbage bins,suitcases and other unusual obstacles.There were a total of 1441 obstacles.It showed that an average True Positive Rate of 94.15% and an average False Alarm Rate of 7.58%.The result shows that the algorithm can effectively detect small obstacles in the parking lot,whether moving or static,and for obstacles with unknown appearance,with certain accuracy and robustness. |
